AI-woordenlijst
Het complete woordenboek van kunstmatige intelligentie
Attribute Inference Attack
Attack where an adversary attempts to infer sensitive attributes not present in the training data from the model's predictions. This attack exploits the implicit correlations learned by the model to reveal private information about individuals.
Shadow Model Attack
Attack where the adversary trains alternative models on synthetic data to mimic the behavior of the target model. These shadow models allow generating training examples to build an effective attack classifier.
Privacy Leak Quantification
Systematic methods for measuring and evaluating the amount of private information disclosed by a machine learning model. These metrics help quantify leak risks and assess the effectiveness of protection mechanisms.
Adversarial Privacy Defense
Proactive defense techniques that incorporate privacy constraints directly into the model's training objective. These methods simultaneously optimize the model's performance and its resistance to inference attacks.
Knowledge Distillation for Privacy
Technique where a private teacher model is used to train a public student model, transferring knowledge while masking sensitive information. This approach reduces the final model's ability to memorize specific details of the training data.
Privacy-Aware Model Design
Architectural design principles integrating privacy protection mechanisms from the model design stage. This approach includes limiting model capacity, adding regularization, and designing less informative outputs.
Model Extraction Attack
Attack where an adversary attempts to replicate or steal a proprietary model by querying its predictions and training a substitute model. This attack can also reveal information about the original training data.